We covered some basic plots previously, but we are going to expand the ability to customize these basic graphics first.
We covered some basic plots previously, but we are going to expand the ability to customize these basic graphics first.
library(readr) death = read_csv( "http://johnmuschelli.com/intro_to_r/data/indicatordeadkids35.csv") death[1:2, 1:5]
# A tibble: 2 x 5
X1 `1760` `1761` `1762` `1763`
<chr> <dbl> <dbl> <dbl> <dbl>
1 Afghanistan NA NA NA NA
2 Albania NA NA NA NA
colnames(death)[1] = "country" death[1:2, 1:5]
# A tibble: 2 x 5
country `1760` `1761` `1762` `1763`
<chr> <dbl> <dbl> <dbl> <dbl>
1 Afghanistan NA NA NA NA
2 Albania NA NA NA NA
library(dplyr) sweden = death %>% filter(country == "Sweden") %>% select(-country) year = as.numeric(colnames(sweden)) plot(as.numeric(sweden) ~ year)
Set within most plots in the base 'graphics' package:
The y-axis label isn't informative, and we can change the label of the y-axis using ylab (xlab for x), and main for the main title/label.
plot(as.numeric(sweden) ~ year,
ylab = "# of deaths per family", main = "Sweden", type = "l")
Let's drop any of the projections and keep it to year 2012, and change the points to blue.
plot(as.numeric(sweden) ~ year,
ylab = "# of deaths per family", main = "Sweden",
xlim = c(1760,2012), pch = 19, cex=1.2,col="blue")
You can also use the subset argument in the plot() function, only when using formula notation:
plot(as.numeric(sweden) ~ year,
ylab = "# of deaths per family", main = "Sweden",
subset = year < 2015, pch = 19, cex=1.2,col="blue")
After reshaping the data to long, we can plot the data with one data.frame:
library(tidyr) long = gather(death, key = year, value = deaths, -country) long = long %>% filter(!is.na(deaths)) head(long)
# A tibble: 6 x 3
country year deaths
<chr> <chr> <dbl>
1 Sweden 1760 2.207555
2 United Kingdom 1760 2.195995
3 Sweden 1761 2.300089
4 United Kingdom 1761 2.347105
5 Sweden 1762 2.785200
6 United Kingdom 1762 2.320127
class(long$year)
[1] "character"
long$year = as.numeric(long$year)
swede_long = long %>% filter(country == "Sweden") plot(deaths ~ year, data = swede_long)
ggplot2 is a package of plotting that is very popular and powerful (using the grammar of graphics). qplot ("quick plot"), similar to plot
library(ggplot2) qplot(x = year, y = deaths, data = swede_long)
The generic plotting function is ggplot, which uses aesthetics:
ggplot(data, aes(args))
g = ggplot(data = swede_long, aes(x = year, y = deaths))
g is an object, which you can adapt into multiple plots!
Common aesthetics:
If you set these in aes, you set them to a variable. If you want to set them for all values, set them in a geom.
geom?g on it's own can't be plotted, we have to add layers, usually with geom_ commands:
geom_point - add pointsgeom_line - add linesgeom_density - add a density plotgeom_histogram - add a histogramgeom_smooth - add a smoothergeom_boxplot - add a boxplotsgeom_bar - bar chartsgeom_tile - rectangles/heatmapsYou "add" things to a plot with a + sign (not pipe!). If you assign a plot to an object, you must call print to print it.
gpoints = g + geom_point(); print(gpoints) # one line for slides
Otherwise it prints by default - this time it's a line
g + geom_line()
You can add multiple geoms:
g + geom_line() + geom_point()
Let's add a smoother through the points:
g + geom_line() + geom_smooth()
If we want a plot with new data, call ggplot again. Group plots by country using colour:
sub = long %>% filter(country %in%
c("United States", "United Kingdom", "Sweden",
"Afghanistan", "Rwanda"))
g = ggplot(sub, aes(x = year, y = deaths, colour = country))
g + geom_line()
Let's remove the legend using the guide command:
g + geom_line() + guides(colour = FALSE)
ggplot(long, aes(x = year, y = deaths)) + geom_boxplot()
For different plotting per year - must make it a factor - but x-axis is wrong!
ggplot(long, aes(x = factor(year), y = deaths)) + geom_boxplot()
geom_jitter plots points "jittered" with noise so not overlappingsub_year = long %>% filter( year > 1995 & year <= 2000) ggplot(sub_year, aes(x = factor(year), y = deaths)) + geom_boxplot(outlier.shape = NA) + # don't show outliers - will below geom_jitter(height = 0)
A facet will make a plot over variables, keeping axes the same (out can change that):
sub %>% ggplot(aes(x = year, y = deaths)) + geom_line() + facet_wrap(~ country)
sub %>% ggplot(aes(x = year, y = deaths)) + geom_line() + facet_wrap(~ country, ncol = 1)
You can use facets in qplot
qplot(x = year, y = deaths, geom = "line", facets = ~ country, data = sub)
You can also do multiple factors with + on the right hand side
sub %>% ggplot(aes(x = year, y = deaths)) + geom_line() + facet_wrap(~ country + x2 + ... )
xlab/ylab - functions to change the labels; ggtitle - change the titleq = qplot(x = year, y = deaths, colour = country, data = sub,
geom = "line") +
xlab("Year of Collection") + ylab("Deaths /100,000") +
ggtitle("Mortality of Children over the years",
subtitle = "not great")
q
?theme_bw - for ggthemesq + theme_bw()
theme - global or specific elements/increase text sizeq + theme(text = element_text(size = 12), title = element_text(size = 20))
q = q + theme(axis.text = element_text(size = 14),
title = element_text(size = 20),
axis.title = element_text(size = 16),
legend.position = c(0.9, 0.8)) +
guides(colour = guide_legend(title = "Country"))
q
transparent_legend = theme(legend.background = element_rect(
fill = "transparent"),
legend.key = element_rect(fill = "transparent",
color = "transparent") )
q + transparent_legend
We can do histograms again using hist. Let's do histograms of death rates over the years:
hist(sub$deaths, breaks = 200)
qplot(x = deaths, fill = factor(country),
data = sub, geom = c("histogram"))
Alpha refers to the opacity of the color, less is more opaque
qplot(x = deaths, fill = country, data = sub,
geom = c("histogram"), alpha=.7)
We cold also do densities:
qplot(x= deaths, fill = country, data = sub,
geom = c("density"), alpha= .7)
colour not fill:qplot(x = deaths, colour = country, data = sub,
geom = c("density"), alpha= .7)
You can take off the lines of the bottom like this
ggplot(aes(x = deaths, colour = country), data = sub) + geom_line(stat = "density")
qplot(x = year, y = deaths, colour = country,
data = long, geom = "line") + guides(colour = FALSE)
Let's try to make it different like base R, a bit. We use tile for the geometric unit:
qtile = qplot(x = year, y = country, fill = deaths, data = sub,
geom = "tile") + xlim(1990, 2005) + guides(colour = FALSE)
scale_fill_gradient let's us change the colors for the fill:
qtile + scale_fill_gradient( low = "blue", high = "red")
Let's try to make it different like base R, a bit. We use tile for the geometric unit:
sub$cat = cut(sub$deaths, breaks = c(0, 1, 2, max(sub$deaths))) qplot(x = year, y = country, fill = cat, data = sub, geom = "tile") + guides(colour = FALSE)
## Stacked Bar Charts
cars = read_csv(
"http://johnmuschelli.com/intro_to_r/data/kaggleCarAuction.csv")
counts <- table(cars$IsBadBuy, cars$VehicleAge)
barplot(counts, main="Car Distribution by Age and Bad Buy Status",
xlab="Vehicle Age", col=c("darkblue","red"),
legend = rownames(counts))
prop.table allows you to convert a table to proportions (depends on margin - either row percent or column percent)
## Use percentages (column percentages)
barplot(prop.table(counts, 2),
main = "Car Distribution by Age and Bad Buy Status",
xlab="Vehicle Age", col=c("darkblue","red"),
legend = rownames(counts))
ggplot(aes(fill = factor(IsBadBuy), x = VehicleAge),
data = cars) + geom_bar()
perc = cars %>% group_by(IsBadBuy, VehicleAge) %>% tally() %>% ungroup head(perc)
# A tibble: 6 x 3
IsBadBuy VehicleAge n
<int> <int> <int>
1 0 0 2
2 0 1 2969
3 0 2 7942
4 0 3 14601
5 0 4 15149
6 0 5 11061
perc_is_bad = perc %>%
group_by(VehicleAge) %>% mutate(perc = n / sum(n))
ggplot(aes(fill = factor(IsBadBuy),
x = VehicleAge,
y = perc),
data = perc_is_bad) + geom_bar(stat = "identity")
## Each Bar adds to 1 for bad buy or not
perc_yr = perc %>%
group_by(IsBadBuy) %>% mutate(perc = n / sum(n))
ggplot(aes(fill = factor(VehicleAge),
x = IsBadBuy,
y = perc),
data = perc_yr) + geom_bar(stat = "identity")
Useful links:
Using the beside argument in barplot, you can get side-by-side barplots.
# Stacked Bar Plot with Colors and Legend
barplot(counts, main="Car Distribution by Age and Bad Buy Status",
xlab="Vehicle Age", col=c("darkblue","red"),
legend = rownames(counts), beside=TRUE)
By default, R displays plots in a separate panel. From there, you can export the plot to a variety of image file types, or copy it to the clipboard.
However, sometimes its very nice to save many plots made at one time to one pdf file, say, for flipping through. Or being more precise with the plot size in the saved file.
R has 5 additional graphics devices: bmp(), jpeg(), png(), tiff(), and pdf()
The syntax is very similar for all of them:
pdf("filename.pdf", width=8, height=8) # inches
plot() # plot 1
plot() # plot 2
# etc
dev.off()
Basically, you are creating a pdf file, and telling R to write any subsequent plots to that file. Once you are done, you turn the device off. Note that failing to turn the device off will create a pdf file that is corrupt, that you cannot open.
These are one of my favorite plots. They are way more informative than the barchart + antenna…
boxplot(weight ~ Diet, data=ChickWeight, outline=FALSE) points(ChickWeight$weight ~ jitter(as.numeric(ChickWeight$Diet),0.5))
Formulas have the format of y ~ x and functions taking formulas have a data argument where you pass the data.frame. You don't need to use $ or referencing when using formulas:
boxplot(weight ~ Diet, data=ChickWeight, outline=FALSE)
R relies on color 'palettes'.
palette("default")
plot(1:8, 1:8, type="n")
text(1:8, 1:8, lab = palette(), col = 1:8)
The default color palette is pretty bad, so you can try to make your own.
palette(c("darkred","orange","blue"))
plot(1:3,1:3,col=1:3,pch =19,cex=2)
It's actually pretty hard to make a good color palette. Luckily, smart and artistic people have spent a lot more time thinking about this. The result is the RColorBrewer package
RColorBrewer::display.brewer.all() will show you all of the palettes available. You can even print it out and keep it next to your monitor for reference.
The help file for brewer.pal() gives you an idea how to use the package.
You can also get a "sneak peek" of these palettes at: http://colorbrewer2.org/ . You would provide the number of levels or classes of your data, and then the type of data: sequential, diverging, or qualitative. The names of the RColorBrewer palettes are the string after 'pick a color scheme:'
palette("default")
plot(weight ~ Time, data= ChickWeight, pch = 19, col = Diet)
library(RColorBrewer) palette(brewer.pal(5,"Dark2")) plot(weight ~ Time, data=ChickWeight, pch = 19, col = Diet)
library(RColorBrewer)
palette(brewer.pal(5,"Dark2"))
plot(weight ~ jitter(Time,amount=0.2),data=ChickWeight,
pch = 19, col = Diet,xlab="Time")
The legend() command adds a legend to your plot. There are tons of arguments to pass it.
x, y=NULL: this just means you can give (x,y) coordinates, or more commonly just give x, as a character string: "top","bottom","topleft","bottomleft","topright","bottomright".
legend: unique character vector, the levels of a factor
pch, lwd: if you want points in the legend, give a pch value. if you want lines, give a lwd value.
col: give the color for each legend level
palette(brewer.pal(5,"Dark2"))
plot(weight ~ jitter(Time,amount=0.2),data=ChickWeight,
pch = 19, col = Diet,xlab="Time")
legend("topleft", paste("Diet",levels(ChickWeight$Diet)),
col = 1:length(levels(ChickWeight$Diet)),
lwd = 3, ncol = 2)
circ = read_csv("http://johnmuschelli.com/intro_to_r/data/Charm_City_Circulator_Ridership.csv")
palette(brewer.pal(7,"Dark2"))
dd = factor(circ$day)
plot(orangeAverage ~ greenAverage, data=circ,
pch=19, col = as.numeric(dd))
legend("bottomright", levels(dd), col=1:length(dd), pch = 19)
dd = factor(circ$day, levels=c("Monday","Tuesday","Wednesday",
"Thursday","Friday","Saturday","Sunday"))
plot(orangeAverage ~ greenAverage, data=circ,
pch=19, col = as.numeric(dd))
legend("bottomright", levels(dd), col=1:length(dd), pch = 19)
We can do histograms again using hist. Let's do histograms of weight at all time points for the chick's weights. We reiterate how useful these are to show your data.
hist(ChickWeight$weight, breaks = 20)
qplot(x = weight,
fill = factor(Diet),
data = ChickWeight,
geom = c("histogram"))
Alpha refers tot he opacity of the color, less is
qplot(x = weight, fill = Diet, data = ChickWeight,
geom = c("histogram"), alpha=.7)
We cold also do densities
qplot(x= weight, fill = Diet, data = ChickWeight,
geom = c("density"), alpha= .7)
qplot(x= weight, colour = Diet, data = ChickWeight,
geom = c("density"), alpha=.7)
ggplot(aes(x= weight, colour = Diet), data = ChickWeight) + geom_density(alpha=.7)
You can take off the lines of the bottom like this
ggplot(aes(x = weight, colour = Diet), data = ChickWeight) + geom_line(stat = "density")
We can make a spaghetti plot by telling ggplot we want a "line", and each line is colored by Chick.
qplot(x=Time, y=weight, colour = Chick,
data = ChickWeight, geom = "line")
In ggplot2, if you want separate plots for something, these are referred to as facets.
qplot(x = Time, y = weight, colour = Chick,
facets = ~Diet, data = ChickWeight, geom = "line")
We can turn off the legend (referred to a "guide" in ggplot2). (Note - there is different syntax with the +)
qplot(x=Time, y=weight, colour = Chick,
facets = ~ Diet, data = ChickWeight,
geom = "line") + guides(colour=FALSE)
ggplot(aes(x = Time, y = weight, colour = Chick),
data = ChickWeight) + geom_line() +
facet_wrap(facets = ~Diet) + guides(colour = FALSE)
Let's try this out on the childhood mortality data used above. However, let's do some manipulation first, by using gather on the data to convert to long.
library(tidyr) long = death long$state = rownames(long) long = long %>% gather(year, deaths, -state) head(long, 2)
# A tibble: 2 x 3 state year deaths <chr> <chr> <chr> 1 1 country Afghanistan 2 2 country Albania
Let's also make the year numeric, as we did above in the stand-alone year variable.
library(stringr)
library(dplyr)
long$year = long$year %>% str_replace("^X", "") %>% as.numeric
long = long %>% filter(!is.na(deaths))
qplot(x = year, y = deaths, colour = state,
data = long, geom = "line") + guides(colour = FALSE)
Let's try to make it different like base R, a bit. We use tile for the geometric unit:
qplot(x = year, y = state, colour = deaths,
data = long, geom = "tile") + guides(colour = FALSE)